Moonshot AI models and APIs

About Moonshot AI

Moonshot AI's Kimi K3 is the long-context lane of the text catalogue. It publishes a single chat task and accepts the OpenAI chat shape.

Kimi K3 takes six fields, a 1,000,000-token window and replies of up to 131,072 tokens, and it accepts image content blocks alongside text, which few chat endpoints here do.

Reach for K3 when the input is the large part: a long transcript, a codebase excerpt, or a scene that includes images. A million-token window only earns its cost once you actually fill it — for short, frequent turns a Flash-tier model elsewhere in the catalogue finishes sooner and bills less. Check the filtering tier on the family page rather than assuming it follows the brand.